Bursty traffic over CDMA: predictive MAI temporal structure, rate control and admission control

We study data communications for the downlink in code-division-multiple-access (CDMA) networks. The focus of this paper is to exploit the predictive temporal structure of the multi-access interference (MAI) for adaptive resource allocation, particularly for rate control and admission control. We also investigate the impact of fading and traffic burstiness on the system performance. We first present our result that when the input data traffic corresponds to multiple ON-OFF heavy tailed sources, the MAI process is asymptotically self-similar (with Hurst parameter 1/2 < H < 1). Thus the MAI has a nontrivial predictive temporal structure that enables accurate interference prediction. We exploit the predictive MAI structure to construct a multiple time-scale interference predictor. Rate adaptation is carried out based on the predicted interference level. Our numerical results show that this rate control scheme achieves significantly better performance than that using the packet-level MAI prediction only. We also devise a joint rate control and admission control scheme. To this end, we propose a sliding observation window scheme that has a two-tier flavor: Each observation window is divided into many time slots, rate control based on the interference prediction is conducted in each slot, and the corresponding throughput in the observation window is used for admission control accordingly. keywords: CDMA, multi-access interference, rate control, admission control, heavy-tailed, self-similar, long-range dependent. ∗Part of this paper was presented in IEEE INFOCOM 2002, New York, NY. This research is supported in part by National Science Foundation grants ANI-0208135, ANI-0207728, and an Intel Research Council grant. †J. Zhang and M. Hu are with the Department of Electrical Engineering, Arizona State University, Tempe, AZ 85287 (e-mail: [email protected]; [email protected]). ‡N. Shroff is with the School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907 (e-mail: [email protected]).